Re: [PATCH 5/8] mm: vmscan: replace shrink_node() loop with a retry jump

[Roman Gushchin <guro@fb.com>]

On Tue, Oct 22, 2019 at 05:42:49PM -0400, Johannes Weiner wrote:
> On Tue, Oct 22, 2019 at 07:56:33PM +0000, Roman Gushchin wrote:
> > On Tue, Oct 22, 2019 at 10:48:00AM -0400, Johannes Weiner wrote:
> > > -			/* Record the group's reclaim efficiency */
> > > -			vmpressure(sc->gfp_mask, memcg, false,
> > > -				   sc->nr_scanned - scanned,
> > > -				   sc->nr_reclaimed - reclaimed);
> > > -
> > > -		} while ((memcg = mem_cgroup_iter(root, memcg, NULL)));
> > > +		reclaimed = sc->nr_reclaimed;
> > > +		scanned = sc->nr_scanned;
> > > +		shrink_node_memcg(pgdat, memcg, sc);
> > >  
> > > -		if (reclaim_state) {
> > > -			sc->nr_reclaimed += reclaim_state->reclaimed_slab;
> > > -			reclaim_state->reclaimed_slab = 0;
> > > -		}
> > > +		shrink_slab(sc->gfp_mask, pgdat->node_id, memcg,
> > > +			    sc->priority);
> > >  
> > > -		/* Record the subtree's reclaim efficiency */
> > > -		vmpressure(sc->gfp_mask, sc->target_mem_cgroup, true,
> > > -			   sc->nr_scanned - nr_scanned,
> > > -			   sc->nr_reclaimed - nr_reclaimed);
> > > +		/* Record the group's reclaim efficiency */
> > > +		vmpressure(sc->gfp_mask, memcg, false,
> > > +			   sc->nr_scanned - scanned,
> > > +			   sc->nr_reclaimed - reclaimed);
> > 
> > It doesn't look as a trivial change. I'd add some comments to the commit message
> > why it's safe to do.
> 
> It's an equivalent change - it's just really misleading because the
> +++ lines are not the counter-part of the --- lines here!
> 
> There are two vmpressure calls in this function: one against the
> individual cgroups, and one against the tree. The diff puts them
> adjacent here, but the counter-part for the --- lines is here:
> 
> > > +	/* Record the subtree's reclaim efficiency */
> > > +	vmpressure(sc->gfp_mask, sc->target_mem_cgroup, true,
> > > +		   sc->nr_scanned - nr_scanned,
> > > +		   sc->nr_reclaimed - nr_reclaimed);
> 
> And the counter-part to the +++ lines is further up (beginning of the
> quoted diff).
> 

Ah, ok, got it. You were right in the foreword, indentation change
diffs are hard to read.

Thanks for the explanation!

Reviewed-by: Roman Gushchin <guro@fb.com>